基于微分几何的遥感影像上线状地物提取方法

Linear Feature Extraction from Remote Sensing Images Based on Differential Geometry

自动识别遥感影像上的目标、快速淮确地获取所需信息是遥感数据处理的主要发展方向。通过研究微分几何理论中曲面的一些基本性质,本文介绍了一种基于微分几何的线状地物提取方法,首先在局部区域内拟合一个二次曲面函数,然后通过该函数来估算灰度曲面的梯度和曲率,设定合适的梯度和曲率阈值来检测线状地物,最后进行后处理,消除许多噪声颗粒及小块区域。实验表明,该方法对于具有一定宽度的线状地物的检测具有较好的效果。

Automatic recognition and acquisition of the objective information from remote sensing images is a main development direction of remote sensing data processing. Through studying some fundamental properties of the curved surface in differential geometry , a method of linear feature extraction based on differential geometry is introduced. First, a conicoid function of the curved surface function is simulated in a given local area. And by this function, the gradient and curvature of the gray curved surface are estimated. Suitable gradient and curvature thresholds are established to detect linear features. Finally, post-processing is performed to remove a lot of noise particles and small areas. The experiments indicate that this method is effective in the extraction of linear features with a certain width.